Prediction of Multidrug-Resistant Tuberculosis Using Machine Learning Algorithms in SWAT, Pakistan.

Area of Science:

• Medical Informatics
• Public Health
• Machine Learning

Background:

• Multidrug-resistant tuberculosis (MDR-TB) presents a significant global health challenge.
• Early and accurate identification of MDR-TB is crucial for effective disease management and control.
• Machine learning (ML) offers potential for improving MDR-TB diagnosis and understanding contributing factors.

Purpose of the Study:

• To apply ML feature selection (FS) algorithms for identifying and diagnosing MDR-TB.
• To identify key factors and symptoms associated with MDR-TB infection.
• To evaluate the performance of various ML models for MDR-TB classification.

Main Methods:

• Analysis of a case-control dataset using ML algorithms: Random Forest (RF), k-Nearest Neighbors (KNN), Support Vector Machine (SVM), Logistic Regression, LASSO, Artificial Neural Networks (ANNs), and Decision Trees.
• Feature selection was employed to identify the most impactful variables for MDR-TB status.
• Model performance was assessed using accuracy, sensitivity, and specificity.

Main Results:

• Close contacts of MDR-TB patients, smoking, depression, prior TB history, improper treatment, and treatment interruptions were identified as significant risk factors.
• Key symptoms associated with MDR-TB include weight loss, chest pain, hemoptysis, and fatigue.
• Support Vector Machine (SVM) and Random Forest (RF) demonstrated superior performance in classifying patients based on accuracy, sensitivity, and specificity.

Conclusions:

• ML algorithms, particularly SVM and RF, are effective tools for the identification and diagnosis of MDR-TB.
• Understanding the identified risk factors and symptoms can aid in early detection and targeted interventions for MDR-TB.
• This study provides valuable insights for developing improved strategies for MDR-TB management and prevention.